Inspiration
Suppose this is another Sunday afternoon, you have a lot of assignments, or housework, maybe boring of songs on billboard and want to try sth new. With our Magic Music Maker, you don’t need to be online and it is possible that you enjoy free music at home!
What it does
We built a music generator by modifying a famous face generating project using DCGANs machine learning model, and we designed two approach to encode/decode music in to pictures.
How we built it
We approach music completion in three steps. We first interpreted musics as matrices using matlab and python in two different approaches respectively. Then we run sampling on our input to satisfy the input of model. Train the DCGAN model and test results. Modify parameters in the whole process to make specific design for our music matrix.
Challenges we ran into
We were not very familiar with Github commands first, and then spent lots of time on configuring our environment for machine learning/computer vision. Sampling is a tricky part, where we need to design a specific encoding/decoding system to satisfy the I/O of the GAN model.
Accomplishments that we're proud of
It was our first time attending a Hachathon and finally we made it. What's more exciting, we gained very unforgettable memory, such as staying up all night coding and rest in sleep bags,...
What we learned
Github commands, conversion from music to matrix, better understanding of tensorflow and DCGAN model
What's next for GANs N' Roses - Magic Music Maker
Try to train a better model (to generate better quality music) and apply in Chrome Extensions, Android apps.
Log in or sign up for Devpost to join the conversation.